NVIDIA RTX PRO 4500 Blackwell Server Edition vGPU profiles: types, hypervisors, desktops

vGPU profiles on the RTX PRO 4500 Server Edition

The NVIDIA RTX PRO 4500 Blackwell Server Edition is shared between virtual machines in two ways: time-sliced vGPUs that take turns on the whole 32 GB card, and MIG-backed vGPUs that sit on one of the card's two 16 GB MIG halves or on the full-card instance. NVIDIA's compute types run from DC-32C, one VM per card, to DC-8C, four VMs per card, and for office desktops NVIDIA's vPC sizing guide counts 16 users per card on 2 GB profiles. The card needs vGPU 20, a branch NVIDIA supports until March 2027.

Desktops from NVIDIA's vPC sizing guide (updated 19 August 2026); compute types from NVIDIA AI Enterprise's Blackwell vGPU types reference (updated 2 September 2026); per-card counts for MIG-backed types are our reading of that table, host totals our arithmetic. All read on 24 October 2026.

The type names follow NVIDIA's scheme for Blackwell. DC-16C is a 16 GB compute vGPU that shares the whole card in turns; DC-1-16C is a 16 GB compute vGPU on one MIG slice. The final letter is the series, and the series sets the licence: C for compute under NVIDIA AI Enterprise, Q for virtual workstations, B for virtual PCs and A for virtual applications. The card-level specifications, from the 165 W board power to the media engines per MIG half, are in our guide to the RTX PRO 4500 Blackwell Server Edition.

Time-sliced vGPU for Compute: DC-32C, DC-16C and DC-8C

Time-sliced vGPUs divide the frame buffer into fixed shares and give each VM the whole GPU's compute in turns. NVIDIA AI Enterprise's reference of Blackwell vGPU types, updated on 2 September 2026, lists three time-sliced compute types for the card, with the same maximum in equal-size and mixed-size mode.

NVIDIA AI Enterprise, Blackwell Architecture vGPU Types, table "Time-Sliced NVIDIA vGPU for Compute for NVIDIA RTX PRO 4500 SE 32 GB", updated 2 September 2026, read 24 October 2026. Required licence edition: NVIDIA AI Enterprise.

There is no compute type below 8 GB on this card, so four VMs is the ceiling for time-sliced compute. DC-32C gives one VM the whole card, which is what an inference VM serving one model needs when it should still move between hosts like any other vGPU VM. DC-16C suits two development VMs with mid-size models, and DC-8C four small notebooks or test VMs.

How much of a vGPU's frame buffer a model can use follows our users-per-card rule: 90 per cent of the memory the driver reports, minus 3 GiB for the runtime. On a whole card that leaves 25.7 GiB for weights and cache, the figure we use for eight cards in our article on a server with eight RTX PRO 4500 Server Edition. Applied to half the card, the same rule leaves about 11 GiB by our estimate, enough for a 7B or 8B model in FP8 with a modest cache. Check the frame buffer the guest driver reports with nvidia-smi inside the VM before you plan a model on a 16 GB vGPU.

MIG-backed vGPU on the two 16 GB halves

NVIDIA's MIG user guide, updated on 13 October 2026, gives the RTX PRO 4500 Blackwell up to two GPU instances, and NVIDIA AI Enterprise's MIG-backed vGPU page states that the Server Edition "supports up to two MIG slices per GPU". A MIG-backed vGPU on one slice has its own memory and compute blocks, so a busy VM on one half does not slow the VM on the other.

NVIDIA AI Enterprise, Blackwell Architecture vGPU Types, table "MIG-Backed NVIDIA vGPU for Compute for NVIDIA RTX PRO 4500 SE 32 GB" (column "Maximum vGPUs per GPU"), updated 2 September 2026, read 24 October 2026. Per-card count: our reading, the listed maximum times the number of instances of that size.

NVIDIA's table names instance profiles such as MIG 2g.16gb and MIG 1g.8gb, which do not appear in the MIG user guide; the guide lists 1g.16gb and 2g.32gb for this card, each with variants such as +gfx. We read the other labels as time-sliced vGPUs inside those two instance sizes, the same reading our guide to RTX PRO 6000 Server Edition vGPU profiles applies to the larger card. The frame buffer sums fit: two DC-1-8C fill one 16 GB half. Check the counts on your own host before you fix a layout.

Three types are time-sliced inside a MIG instance in this reading: DC-2-16C and DC-2-8C share the full-card instance, and DC-1-8C shares a half. NVIDIA's vGPU 20.0 release notes list "Time-sliced, MIG-backed vGPUs within a GPU instance" as new, and on vSphere they list this feature under VCF 9.1. On ESXi 8.0 and VCF 9.0, plan on the types that fill an instance: DC-1-16C and DC-2-32C. NVIDIA's vGPU user guide marks debuggers and profilers as unsupported in the "MIG-backed, time-sliced" mode on this card, while time-sliced and plain MIG-backed vGPUs support both. Release 20.2 adds several MIG-backed vGPUs in one VM on VCF 9.1 or later.

Graphics profiles for VDI and virtual workstations

For office desktops and CAD workstations the card uses B-series and Q-series profiles. NVIDIA's vPC sizing guide, updated on 19 August 2026, calls the RTX PRO 4500 Blackwell Server Edition "the primary recommended GPU for NVIDIA vPC". It gives 16 users per card on 2 GB profiles and 10 on 3 GB profiles, notes that 1 GB profiles are not supported on Blackwell, and counts eight cards and 128 users per server in a table that "reflects maximum supported density rather than a recommended deployment point". NVIDIA's vWS selection guide calls the card "the recommended GPU for entry-level vWS deployments", and NVIDIA's RTX vWS sizing guide, also updated on 19 August 2026, puts 6 to 8 light users per card on the DC-4Q profile and 3 to 6 medium users on DC-4Q or DC-8Q; for heavy users it lists only RTX PRO 6000 profiles.

NVIDIA's technical blog on the card and vGPU 20, published on 22 April 2026, runs three VMs on the two MIG halves in a vSphere demo: one with a 16Q profile on the first instance, and a 4Q and a 3B profile together on the second. It names the profile nvidia_rtx_pro_4500_blackwell_dc-1-4q as "a 4 GB profile" and says that NVIDIA's "Mixed Size" setting lets different profile sizes share one MIG instance.

The full Q-series, B-series and A-series tables for the card are in section 9.1.12 of NVIDIA's vGPU user guide, release 20.0 to 20.2 of 30 September 2026. We could not open those tables in full on 24 October 2026, so this article gives no graphics profile list beyond the figures above. Take the maximum number of each graphics type per card from that section, or from nvidia-smi vgpu output on a host with the vGPU manager installed, before you size an estate. Our guide to VDI GPU sizing shows how to scale hosts to 1,000 desktops once the per-card figure is fixed.

We build servers with RTX PRO 4500 Server Edition cards for VDI and compute VMs and quote the vPC, vWS or NVIDIA AI Enterprise licences with the hardware; send us your user counts per type and your hypervisor version, and our team replies with a host count and a configuration.

vGPU 20 only: branch support dates for a host bought now

As of 24 October 2026, NVIDIA's list of GPUs supported by vGPU software, updated on 2 October 2026, shows the RTX PRO 4500 Blackwell Server Edition with release 20.0 as the first supported release, status "Full Support", and the last supported branch as "TBD". Releases 19.x do not support it. NVIDIA's vGPU documentation index gives the branch dates.

NVIDIA vGPU software documentation index (release branches), and NVIDIA's list of GPUs supported by vGPU software, updated 2 October 2026; both read 24 October 2026. vGPU 19 does not list the RTX PRO 4500 Server Edition.

A host bought today therefore starts on a production branch with about five months of support left. Plan the move to the next branch during the first year in service, and test it on one host before the rest. vGPU manager and guest driver have to fit together, so the update touches the hypervisor host and every guest image. The long-term branch, vGPU 19, would give a longer quiet period, but only the RTX PRO 6000 Blackwell Server Edition among the RTX PRO cards is on it.

Hypervisors that list the card

NVIDIA's product support matrices for the current release, updated on 29 September 2026, list the card on these hypervisors.

NVIDIA vGPU product support matrices for VMware vSphere, Red Hat Enterprise Linux with KVM, Ubuntu, XenServer and Windows Server, updated 29 September 2026, read 24 October 2026; MIG-backed and Windows Server 2022 notes from NVIDIA's validated-platforms pages for Windows Server and XenServer (updated 29 September 2026), read 25 October 2026. Red Hat OpenShift Virtualization does not list the card.

On vSphere, footnote 6 of the matrix states that vGPU "requires the vSphere Foundation edition" of ESXi or a vSphere Enterprise Plus licence. ESXi 8.0 builds before Update 3g do not list the card at all, so an older host image has to be updated before the cards go in. Our guide to GPUs in VMware vSphere compares passthrough with vGPU on that platform.

Desktops and VMs per eight-card host

Supermicro lists up to eight RTX PRO 4500 Server Edition cards in its 3U, 4U and 5U models and up to six in 2U. NVIDIA's vPC density table assumes eight boards in a 2U server without naming one, so we do not count a 2U chassis as an eight-card option. An eight-card host gives these maximums by NVIDIA's figures and our arithmetic.

NVIDIA vPC sizing guide (19 August 2026) and NVIDIA AI Enterprise Blackwell vGPU types (2 September 2026); per-host totals our arithmetic for eight cards. Graphics maximums are NVIDIA's maximum densities, not recommended deployment points.

The vPC figures are ceilings. NVIDIA's sizing guide says the right profile size depends on the workload and environment and recommends a proof of concept on the target GPU and software; if your desktops need 3 GB instead of 2 GB, a host drops from 128 to 80 desktops. Every vGPU VM also needs processor cores and memory on the host: NVIDIA's certified-systems guide asks for at least six cores per GPU and system memory of at least twice the GPU memory, 48 cores and 512 GB for eight cards, before the VMs' own allocations. With 128 desktops of two or four virtual processors each, the virtual processors set the core count.

Power is the easy part. Eight cards draw 1,320 W at 165 W each, and with two 280 W processors and 15 per cent for the rest of the server we estimate about 2.2 kW at the wall, as in our article on the eight-card server. That stays under the 2.94 kW we plan per 230 V 16 A feed.

Plan spare capacity per profile. A VM moves to another host only where a free vGPU of the same type is left, and NVIDIA's MIG-backed vGPU page notes that live migration between different MIG profiles is not supported. With two hosts of eight cards each, half the users of each pool lose their desktop when one host is down, unless the other host keeps free vGPUs of the same type.

We supply servers with up to eight RTX PRO 4500 Server Edition cards, assembled and tested before delivery, with the vGPU licences on the same quote; describe your desktop and VM mix, and we size the hosts and the spare capacity.

Licences per profile series

The series letter of each type sets the licence. NVIDIA's client licensing guide, updated on 29 September 2026, maps Q-series vGPUs to RTX vWS, B-series to vPC or vWS, and A-series to vApps. It counts one vApps licence for each concurrent user, and for Q-series it states that "only one license for the VM is required, not one license for each vGPU assigned to the VM". C-series compute types require NVIDIA AI Enterprise, which NVIDIA licenses "on a per-GPU basis".

For NVIDIA AI Enterprise support, the 8.2 support matrix lists the RTX PRO 4500 Blackwell Server Edition among its discrete GPUs, with the condition that GPUs "are supported with servers listed on the NVIDIA-Certified Systems". If you need NVIDIA's support for compute VMs, specify a server model that NVIDIA lists as certified with this card. Our page on NVIDIA AI Enterprise and vGPU licences explains the editions, and our guide to the NVIDIA licence server covers DLS and CLS.

When another card fits better

The RTX PRO 4500 Server Edition is the dense choice: many small vGPUs per watt in single-slot cards. Three cases point elsewhere. A VM that needs more than 32 GB of frame buffer, or a model that needs more than one 16 GB half, belongs on the RTX PRO 6000 Blackwell Server Edition, which offers four 24 GB MIG slices and types up to 96 GB and sits on the long-term vGPU 19 branch. An estate that cannot move to a new vGPU branch within a year should also look at the RTX PRO 6000 Server Edition for that reason. For office desktops at 1 GB per user, the L4 still offers that profile size, which Blackwell does not; our L4 vGPU profiles and L40S vGPU profiles list those types, and our comparison of the L40S and RTX PRO 4500 Server Edition weighs 48 against 32 GB per card.

FAQ